Meta’s Muse Spark climbs to number two on OpenCode Go with staggering usage numbers

Meta’s Muse Spark has quietly become one of the most-used AI coding models in the world. The model now ranks second on OpenCode Go’s weekly leaderboard, processing over 120 million sessions and 79 trillion developer tokens across the platform. What Muse Spark actually does Muse Spark emerged from Meta Superintelligence Labs as a purpose-built model for agentic coding. The model operates with a 1 million token context window, which means it can hold roughly the equivalent of several large codebases in its working memory at once. The model series has evolved quickly since its April 2026 debut. Version 1.1 arrived in July, followed by version 1.2 on August 5, 2026, which was co-trained with Muse Code, Meta’s dedicated terminal agent. Version 1.3 landed in September, continuing the rapid iteration cycle. OpenCode Go, the platform where Muse Spark has earned its number-two ranking, operates as a $10-per-month subscription service that gives developers access to various AI coding models optimized for terminal agents. The ranking is based on actual usage rather than synthetic benchmarks. The numbers in context The 120 million sessions metric suggests not just raw throughput but repeated, ...